An Application of Variable Selection Methods to Identify Influential Factors of ADHD Diagnosis in Children

Abstract

In 2011, there was the ADHD-200 Global Competition in which several teams around the world competed to produce a model that best categorized children as having ADHD or not having ADHD based on phenotypic and neu- roimaging data for each child. These data were gathered at eight locations in the United States, the Netherlands, and China. The model each team pro- duced was tested and scored based on prediction accuracy. Interestingly, the team that scored the highest used only the phenotypic data. This team was dis- qualied because it did not use the neuroimaging data, so its model was never published. The results of the competition show that there is a relationship between phenotypic variables and ADHD diagnosis. The goal of this thesis is to identify which variables most greatly in uence ADHD diagnosis in children using penalized logistic regression. According to the CDC about 11% of United States children between the ages of 4 and 17 are diagnosed with ADHD. It is important that children are correctly diagnosed. This thesis will help identify which factors are infuential in ADHD diagnosis in children.